quantified.uncertainty/squiggle
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Fixes CdfLibrary.js

Browse Source
This commit is contained in:
Roman Galochkin 2020-02-25 13:13:17 +03:00
parent 4fc7723e32
commit 9349930ad8
1 changed files with 165 additions and 169 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

334
src/utility/CdfLibrary.js
View File

@ -1,174 +1,170 @@
const { const {
Cdf, Cdf,
Pdf, Pdf,
ContinuousDistribution, ContinuousDistribution,
ContinuousDistributionCombination, ContinuousDistributionCombination,
scoringFunctions, scoringFunctions,
} = require("@foretold/cdf/lib"); } = require("@foretold/cdf/lib");
const _ = require("lodash"); const _ = require("lodash");
/**
*
* @param xs
* @param ys
* @returns {{ys: *, xs: *}}
*/
function cdfToPdf({ xs, ys }) {
let cdf = new Cdf(xs, ys);
let pdf = cdf.toPdf();
return { xs: pdf.xs, ys: pdf.ys };
}
/** /**
* *
* @param xs * @param xs
* @param ys * @param ys
* @returns {{ys: *, xs: *}} * @returns {{ys: *, xs: *}}
*/ */
function pdfToCdf({ xs, ys }) { function cdfToPdf({ xs, ys }) {
let cdf = new Pdf(xs, ys); let cdf = new Cdf(xs, ys);
let pdf = cdf.toCdf(); let pdf = cdf.toPdf();
return { xs: pdf.xs, ys: pdf.ys }; return { xs: pdf.xs, ys: pdf.ys };
} }
/**
*
* @param sampleCount
* @param vars
* @returns {{ys: *, xs: *}}
*/
function mean(sampleCount, vars) {
let cdfs = vars.map(r => new Cdf(r.xs, r.ys));
let comb = new ContinuousDistributionCombination(cdfs);
let newCdf = comb.combineYsWithMean(sampleCount);
return { xs: newCdf.xs, ys: newCdf.ys };
}
/**
*
* @param sampleCount
* @param predictionCdf
* @param resolutionCdf
*/
function scoreNonMarketCdfCdf(sampleCount, predictionCdf, resolutionCdf, resolutionUniformAdditionWeight=0) {
let toCdf = (r) => (new Cdf(r.xs, r.ys));
let prediction = toCdf(predictionCdf);
if (_.isFinite(resolutionUniformAdditionWeight)){
prediction = prediction.combineWithUniformOfCdf(
{
cdf: toCdf(resolutionCdf),
uniformWeight: resolutionUniformAdditionWeight,
sampleCount
}
);
}
return scoringFunctions.distributionInputDistributionOutputMarketless({
predictionCdf: prediction,
resultCdf: toCdf(resolutionCdf),
sampleCount,
});
}
/**
*
* @param sampleCount
* @param cdf
*/
function differentialEntropy(sampleCount, cdf) {
let toCdf = (r) => (new Cdf(r.xs, r.ys));
return scoringFunctions.differentialEntropy({
cdf: toCdf(cdf),
sampleCount: sampleCount
});
}
/**
*
* @param x
* @param xs
* @param ys
* @returns {number}
*/
function findY(x, { xs, ys }) {
let cdf = new Cdf(xs, ys);
return cdf.findY(x);
}
/** /**
* *
* @param x * @param xs
* @param xs * @param ys
* @param ys * @returns {{ys: *, xs: *}}
* @returns {number[]} */
*/ function pdfToCdf({ xs, ys }) {
function convertToNewLength(n, { xs, ys }) { let cdf = new Pdf(xs, ys);
let dist = new ContinuousDistribution(xs, ys); let pdf = cdf.toCdf();
return dist.convertToNewLength(n); return { xs: pdf.xs, ys: pdf.ys };
} }
/** /**
* *
* @param y * @param sampleCount
* @param xs * @param vars
* @param ys * @returns {{ys: *, xs: *}}
* @returns {number} */
*/ function mean(sampleCount, vars) {
function findX(y, { xs, ys }) { let cdfs = vars.map(r => new Cdf(r.xs, r.ys));
let cdf = new Cdf(xs, ys); let comb = new ContinuousDistributionCombination(cdfs);
return cdf.findX(y); let newCdf = comb.combineYsWithMean(sampleCount);
}
return { xs: newCdf.xs, ys: newCdf.ys };
/** }
*
* @param xs /**
* @param ys *
* @returns {number[]} * @param sampleCount
*/ * @param predictionCdf
function integral({ xs, ys }) { * @param resolutionCdf
if (_.includes(ys, NaN)){ */
return NaN; function scoreNonMarketCdfCdf(sampleCount, predictionCdf, resolutionCdf, resolutionUniformAdditionWeight = 0) {
} let toCdf = (r) => (new Cdf(r.xs, r.ys));
else if (_.includes(ys, Infinity) && _.includes(ys, -Infinity)){ let prediction = toCdf(predictionCdf);
return NaN; if (_.isFinite(resolutionUniformAdditionWeight)) {
} prediction = prediction.combineWithUniformOfCdf(
else if (_.includes(ys, Infinity)){ {
return Infinity; cdf: toCdf(resolutionCdf),
} uniformWeight: resolutionUniformAdditionWeight,
else if (_.includes(ys, -Infinity)){ sampleCount
return -Infinity;
}
let integral = 0;
for (let i = 1; i < ys.length; i++) {
let thisY = ys[i];
let lastY = ys[i - 1];
let thisX = xs[i];
let lastX = xs[i - 1];
if (
_.isFinite(thisY) && _.isFinite(lastY) &&
_.isFinite(thisX) && _.isFinite(lastX)
) {
let sectionInterval = ((thisY + lastY) / 2) * (thisX - lastX);
integral = integral + sectionInterval;
} }
);
}
return integral;
} }
module.exports = { return scoringFunctions.distributionInputDistributionOutputMarketless({
cdfToPdf, predictionCdf: prediction,
pdfToCdf, resultCdf: toCdf(resolutionCdf),
findY, sampleCount,
findX, });
convertToNewLength, }
mean,
scoreNonMarketCdfCdf, /**
differentialEntropy, *
integral, * @param sampleCount
}; * @param cdf
*/
function differentialEntropy(sampleCount, cdf) {
let toCdf = (r) => (new Cdf(r.xs, r.ys));
return scoringFunctions.differentialEntropy({
cdf: toCdf(cdf),
sampleCount: sampleCount
});
}
/**
*
* @param x
* @param xs
* @param ys
* @returns {number}
*/
function findY(x, { xs, ys }) {
let cdf = new Cdf(xs, ys);
return cdf.findY(x);
}
/**
*
* @param x
* @param xs
* @param ys
* @returns {number[]}
*/
function convertToNewLength(n, { xs, ys }) {
let dist = new ContinuousDistribution(xs, ys);
return dist.convertToNewLength(n);
}
/**
*
* @param y
* @param xs
* @param ys
* @returns {number}
*/
function findX(y, { xs, ys }) {
let cdf = new Cdf(xs, ys);
return cdf.findX(y);
}
/**
*
* @param xs
* @param ys
* @returns {number[]}
*/
function integral({ xs, ys }) {
if (_.includes(ys, NaN)) {
return NaN;
} else if (_.includes(ys, Infinity) && _.includes(ys, -Infinity)) {
return NaN;
} else if (_.includes(ys, Infinity)) {
return Infinity;
} else if (_.includes(ys, -Infinity)) {
return -Infinity;
}
let integral = 0;
for (let i = 1; i < ys.length; i++) {
let thisY = ys[i];
let lastY = ys[i - 1];
let thisX = xs[i];
let lastX = xs[i - 1];
if (
_.isFinite(thisY) && _.isFinite(lastY) &&
_.isFinite(thisX) && _.isFinite(lastX)
) {
let sectionInterval = ((thisY + lastY) / 2) * (thisX - lastX);
integral = integral + sectionInterval;
}
}
return integral;
}
module.exports = {
cdfToPdf,
pdfToCdf,
findY,
findX,
convertToNewLength,
mean,
scoreNonMarketCdfCdf,
differentialEntropy,
integral,
};